I also wanted to point out that the comment below does not apply to 10G
ports. They are individually much more expensive to provision.
Antonio
> Each high-performance switch port costs SLAC about $400. However, depending
> on where your machines get installed, there may be an existing switch with
> available ports near (b050, 2nd floor), so in that case you
> end up getting the ports for "free". Not the greatest charge-back model, but
> that's what we have.
>
> Antonio
>
> Yang, Wei wrote Today:
>
>> Hi Antonio,
>>
>> We are looking at ~10 ports. How much per port?
>>
>> regards,
>> Wei Yang | [log in to unmask] | 650-926-3338(O)
>>
>>
>>
>>
>> On Jul 19, 2010, at 12:40 PM, Antonio Ceseracciu wrote:
>>
>>>
>>> ...do you need to procure 1Gb switch ports then? How many?
>>>
>>> Antonio
>>>
>>> Yang, Wei wrote Today:
>>>
>>>> Here is a summary (and my opinion) after talking to several of you on
>>>> Friday about the proof configuration:
>>>>
>>>> 1. Spindles. Both David and Bart, and the Wisc Proof people think that
>>>> more spindles are important. R510 can host 12 data disks while R710 can
>>>> only host 6 (for system and data together). So R510 clearly has an
>>>> advantage. There were suggestion to use 1TB drives instead of 2TB drives
>>>> to save money for CPUs. After carefully examined the config, I think we
>>>> should stay with 2TB drives because doing so will cut disk space by half
>>>> (~50TB) and only give us 8 more cores.
>>>>
>>>> 2. 10Gb. The price quote for Dell PowerConnect 8024 10Gb switch is pretty
>>>> expensive, ~$9400. Someone who previously worked at AGL Tier 2 (who I
>>>> happened to meet on Friday) told me that AGLT2 had bad experience with
>>>> the Dell switches. So I suggest that we go with 1Gb network. This should
>>>> not have negative impact of the data loading and offloading speed as long
>>>> as users have several data transfer streams in parallel. The needs for
>>>> inter-cluster network should be minimized by Proof itself.
>>>>
>>>> 3. The head node may still need to be a 10Gb so that users can use it to
>>>> manually transfer data in and out. I will look for a R710 to hook it to
>>>> the Tier 2's 8024F switch.
>>>>
>>>> 4. DDM site. We discussed of having a DDM site (SRM endpoint) for the
>>>> cluster in order to allow direct DDM transfer in and out, in addition to
>>>> use dq2-get to write directly to the cluster. I think a DDM site is a
>>>> good think to have but I need to talk to the US Tier 3 technical people
>>>> on catalog setup (so that is no guarantee) , and we should remember that
>>>> with a DDM site it is always easy to copy data in but hard to delete
>>>> them.
>>>>
>>>> So i will work with Teri Church on the price of a R710 with 10Gb and
>>>> several R510s with 1Gb unless I hear objections ...
>>>>
>>>> regards,
>>>> Wei Yang | [log in to unmask] | 650-926-3338(O)
>>>>
>>>>
>>>>
>>>>
>>>
>>> __
>>> . __ ____
>>> . /_/\ /_/_/ Antonio Ceseracciu Network
>>> . /_/\_\ /_/ [log in to unmask] Specialist
>>> ./_/_/\_\ \_\ _ SLAC +1 (650) 926 2895 SLAC/SCCS
>>> ._/ \_\ \_\_\
>>
>>
>
> __
> . __ ____
> . /_/\ /_/_/ Antonio Ceseracciu Network
> . /_/\_\ /_/ [log in to unmask] Specialist
> ./_/_/\_\ \_\ _ SLAC +1 (650) 926 2895 SLAC/SCCS
> ._/ \_\ \_\_\
>
__
. __ ____
. /_/\ /_/_/ Antonio Ceseracciu Network
. /_/\_\ /_/ [log in to unmask] Specialist
./_/_/\_\ \_\ _ SLAC +1 (650) 926 2895 SLAC/SCCS
._/ \_\ \_\_\
|